Dear Google AI community,
I’ve implemented a Chatbot with File Search support via API. Since yesterday, we see sporadically bad answers which are based on a specific pdf. Since today, it’s not sporadically anymore. So I guess the role out of gemini-embedding-2 is the reason for the bad results.
Has anyone else encountered a similar issue?
Can we do something in order to improve the answers (switch back to embedding-1?)
Thanks and kind regards,
Philipp